magic_get VS pfr

Maybe this: https://github.com/apolukhin/magic_get

You can access non-static member variables with structured bindings. See for instance magic_get/boostPfr https://github.com/apolukhin/magic_get. Structured bindings will only bind to the accessible members (not private). Magic get works by by finding the number of member variables, converting the struct to a tuple (with a function that is specialized for the number of fields), and finally accessing the members through the tuple (supporting indexing and iteration). It does not grant access to the member names, but it is sufficient for some reflection.

Libraries like magic_get expose the members of an aggregate class/struct to allow writing generic code for things like pretty printing and serialization without anything special done to the class itself. They often rely on structured bindings for the decomposition (*), but find the number of elements via SFINAE on aggregate initialization, as an aggregate type can be initialized only from as many objects as it has members. It would be nicer if you could SFINAE directly on the structured binding itself, as then the type could have user-defined constructors (which aggregates can't). Unfortunately, this is not possible since structured binding is a statement and not an expression. Unless you're using Clang, where the GNU statement expressions extension allows you to do SFINAE on them, as in here.

No RTTI required, but the types are required to be aggregates (no constructors defined). It's possible to count the number of members using SFINAE by trying different numbers of inputs to the aggregate constructor using a type that's castable to anything, and then enumerate the members with a similar trick (or use structured binding to pull them out directly). I think he uses magic_get which is the most popular library for this trick.

take a look at magic_get to get access to all struct members. No idea what your plan is with foo and the preprocessor though.

Neat! Have you considered using magic_get?

It is possible with some hacks https://github.com/apolukhin/magic_get

I actually reference it near the bottom of the article under its original name, magic_get! I was disappointed to discover that this library did it a similar way first before me while researching 😅

This single feature opens a world of new possiblities. For example, it makes implementing "getting the number of fields" trivial. Furthrmore, and much more importantly, it enables turning a struct into a tuple. Currently, this can only be done by enumerating cases (therefore it's not fully generic), as with Boost PFR. By the way, PFR greatly simplifies our codebases, especially for parts with serialization and/or reflection.

The code is relatively short and can be groked with a few coffees: https://github.com/boostorg/pfr/tree/develop/include/boost/pfr ; if you're using C++17 it uses a binary search (https://github.com/boostorg/pfr/blob/develop/include/boost/pfr/detail/fields_count.hpp) to count the number of fields in a struct, by starting by the observation that a likely majorant on the number of fields in a struct is sizeof(the struct) * CHAR_BIT, assuming not too many [[no_unique_address]] tomfooleries. Then once this count is known it's possible to simply map them as a tuple through sheer brute force and destructuring: https://github.com/boostorg/pfr/blob/develop/include/boost/pfr/detail/core17_generated.hpp

I wonder if the c++ approach of boost.pfr would be portable to rust ? It allows reflection on aggregates without needing to annotate anything: https://github.com/boostorg/pfr

It is an 'interesting' meta-programming problem though (wasted many weeks on it myself, fixed a small gcc bug - a 'uniform init' edge case and filed an issue with magic_get Reflecting array members of aggregate structs).